ITHome · Published 2026-09-10
DeepSeek V4.1 Flash Model Goes Live on the National Supercomputing Internet
DeepSeek V4.1 Flash has joined the National Supercomputing Internet’s model service. The report gives a route from its homepage to an API interface and describes changes to input computation and caching.
Open the model service’s API interface
ITHome reported on September 10 that users can sign in to the National Supercomputing Internet site, open Model Services from the homepage and access the DeepSeek V4.1 Flash API interface. The report identifies this new entry point without specifying its plans or rates.
Input computation and cache storage
The reported 552B-parameter MoE matches the official model card. Its causal encoder–decoder separates input and output computation, activating about 8B parameters per token during prefill and 16B during decoding.
The model card reports a global KV cache about one quarter the size of V4 Flash’s and a persistent cache about one eighth the size. These address long inputs and context storage; the ratios are not a quoted discount on the platform’s bills.
Important additions and corrections
Dates below mark changes to our coverage.
- Addition ·
Added the platform’s Model Services entry and model-card context for computation and caching.
Sources and notes
sources中文
Sources and notes
Retained summary and context
Read claims in the context of the original page. Claims remain attributed to their sources. Collection is not proof of the claims or practical outcomes.
Sources and evidence · 2 links
Each link has a purpose and scope. Link counts do not establish independent confirmation; direct support applies only to the statements specified below.
This page keeps the original link and a short account. Follow the links below for the publisher’s full explanation.
Source timeline
- · Found this reading link via ITHome.
Collection record
First discovered: ; added to the library:
Collection and source publication have distinct dates.
Editorial record
Sources reviewed
· Beijing time (UTC+8)
Added the platform’s Model Services entry and model-card context for computation and caching. Original event, discovery and collection dates remain.
Sources and full record
This dates our review of the wording and sources, not a project release or product update.
Full record: before, after and sources (JSON)